[erikdarlingdata]

Claude posting for Erik Darling.

Fixes #6488.

Why

The 3.11.0 release PR (#6481) conflicted because 3.10.0 (#5645) was squashed into main, so dev's commits never reached main's history. #6485 merged main into dev with a merge commit, which cleared the conflicts. The release checklist did not say to use a merge commit, so the next release could repeat the conflict.

Git history for this, from git log:

• 93a67dee6 Release 3.10.0 (Release 3.10.0 #5645) (Merge main into dev so the 3.11.0 release PR has no conflicts #6483): the squash.
• 7d6912c0a Merge main into dev so the 3.11.0 release PR has no conflicts: the sync commit.
• 71bfee79c Merge pull request Merge main into dev (merge commit) so the 3.11.0 release PR has no conflicts #6485: the merge that cleared it.
• 1b6af5187 Merge pull request Release 3.11.0 #6481 from dev: the 3.11.0 release PR.

What changes

• .claude/skills/release-checklist/SKILL.md, step 10 "Commit and PR": one bullet added after the "open the release PR from dev to main" bullet (now line 240). It says to merge the dev to main release PR with Create a merge commit, never squash, and gives the reason with the 3.10.0 and 3.11.0 history. The bullet text is the brief's text, unchanged.

No other line changes. git diff origin/dev...HEAD shows one added line in one file.

Test plan

• plain-english/check.py on SKILL.md: no hits on the new bullet (line 240). The file has 107 hits in existing text. Per the brief, those were left alone.
• Diff is one added line in one file.
• No tests run. The brief says this is a doc edit: build nothing, run no tests. Darling/Darling.Tests/StatementColumnCensusTests.cs mentions the release-checklist skill only in a comment (line 934) and does not read SKILL.md, so no test reads this change.

CHANGELOG

None (release process doc)

For the coordinator

• Nothing to double-check beyond the one-line diff.

[claude]

Verdict: LGTM

Claims

• VERIFIED — "one bullet added after the 'open the release PR from dev to main' bullet (now line 240)" — confirmed via diff hunk arithmetic (@@ -237,6 +237,7 @@, new line is the 4th of 7 context+added lines, landing at 240) and git show HEAD:.claude/skills/release-checklist/SKILL.md line 240
• VERIFIED — "No other line changes" / single-line diff — gh pr view reports additions:1 deletions:0, and gh pr diff shows exactly one added line in one file
• VERIFIED — "The release checklist did not say to use a merge commit, so the next release could repeat the conflict" — grep for squash/merge commit across the pre-existing file (minus the new line) finds zero hits before this change
• VERIFIED — "Darling/Darling.Tests/StatementColumnCensusTests.cs mentions the release-checklist skill only in a comment (line 934) and does not read SKILL.md" — confirmed: line 934 is a /// comment, no other reference to SKILL.md/release-checklist in that file
• UNVERIFIED — "The 3.11.0 release PR (#6481) conflicted because 3.10.0 (#5645) was squashed into main" and "#6485 merged main into dev with a merge commit, which cleared the conflicts" — this checkout is shallow (2 commits visible) and the review's tool budget is scoped to PR 6489 only, so the cited PR/issue history can't be independently confirmed from this tree
• UNVERIFIED — "plain-english/check.py on SKILL.md: no hits on the new bullet... 107 hits in existing text" — no plain-english directory, script, or reference to one exists anywhere in this checkout (grep -r "plain-english" repo-wide is empty); plausibly tooling external to the repo (the brief/coordinator tooling the body also references), but it cannot be confirmed or refuted from the tree

Riskiest lines

1. .claude/skills/release-checklist/SKILL.md:240 (the single added bullet) — it's prose, not code: the only "risk" is the historical claim being wrong or the GitHub merge-method UI not matching, neither of which is checkable from a read-only tree — UNCOVERED, but the PR body's own test-plan item 3 ("No tests run... this is a doc edit") is the one-sentence justification the coverage rule asks for
2. N/A — the diff contains exactly one changed line; there is no second risky line
3. N/A — same reason

Traps near this diff

• ".gitattributes says text=auto eol=crlf... a diff that touches every line of a file is a line-ending conversion until proven otherwise" — does not apply: gh pr diff shows a clean single-line addition with all surrounding context lines unchanged, not a whole-file rewrite (note: the sandbox's working copy of this file locally differs from the committed HEAD by this same one line, pre-existing at session start per the attached gitStatus, unrelated to the PR's actual committed content, which matches gh pr diff exactly)
• "Lite and Darling ship in parity... a fix to one SKU's reader, tool, alert or description without the counterpart's is a finding" — does not apply: SKILL.md is a single shared release-process document already covering both SKUs' steps (e.g. the Linux/Darling-specific section at line 211+ sits in the same file as the Windows steps); this bullet is about the shared dev→main PR step, not a SKU-specific surface

Divergence (blind read vs PR body)

None — the body's account matches the diff on every claim that could be checked from this tree. The two historical-event claims (3.10.0/3.11.0/#6485) and the plain-english/check.py test-plan claim are outside what a read-only checkout of this scope can confirm or refute, but nothing in the tree contradicts them either.

Reviewer checklist

• Confirm the repository's GitHub merge-button settings actually allow "Create a merge commit" (not restricted to squash-only) for PRs into main, since the new instruction assumes that option is available